Revolut cut AI coding costs 90% by self-hosting GLM on 56 GPUs
Revolut was spending up to $10,000 per engineer per month on commercial AI coding tools with no clear return on that spend. By April 2026, the British fintech had found a way to give 2,000 engineers unrestricted access to AI coding tools for a fraction of that cost.
The Setup
Revolut now self-hosts GLM-5.3, the open-weight coding model from Chinese AI lab Z.ai, on 56 NVIDIA B200 GPUs running on US-based infrastructure. The company downloads model weights and runs inference domestically rather than routing requests to Chinese-hosted endpoints.
The blended token cost came out to $0.30 per million tokens. According to David Abitbol, who leads AI operations at Revolut and presented these numbers at the AI Engineering conference, that works out to one-tenth of what the company was previously invoicing with Anthropic and OpenAI.
What Changed for Engineers
Removing token budgets had a knock-on effect beyond raw cost savings. Revolut reports:
- 70% improvement in engineering lead times
- 60% reduction in coding time across the 2,000-engineer sample
- Less time spent searching documentation, managing inboxes, and coordinating dev tasks
The infrastructure targets a time to first token below three seconds. Abitbol noted that latency above five seconds generates engineer complaints. Interactive workloads run at around 200 tokens per second, with throughput available up to 800 tokens per second after 5 pm when GPU demand drops.
Performance Gains Without More Hardware
Between April and August, Revolut pushed request throughput from roughly 3 requests per second to more than 7, while cutting time to first token from over 10 seconds to around 1.6 seconds. That is an 88% performance improvement against a 76% increase in demand. Abitbol attributed this to inference optimization, not additional GPU capacity.
Frontier Models Still in the Mix
Revolut has not fully walked away from commercial providers. Claude Code, Codex, and Gemini remain in use for specialized workloads including image recognition. The self-hosted GLM stack handles most coding tasks.
CEO Nik Storonsky separately said in October that Revolut is developing proprietary AI models trained on the company’s 30 to 40 million daily customer transactions.
“Open source is now a priority at Revolut. The economics are favourable, the billing is flat, and we don’t have any bill shock at the end of the month.” — David Abitbol, AI Operations Lead, Revolut
For any operator running a developer team where AI tool costs are climbing without a clear productivity signal, Revolut’s playbook is worth studying: cap commercial spend first, then find an open-weight model that fits the workload, then optimize inference before reaching for more hardware.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
